CN112631250B - Fault isolation and identification method in nonlinear process based on denoising autoencoder - Google Patents

Fault isolation and identification method in nonlinear process based on denoising autoencoder Download PDF

Info

Publication number
CN112631250B
CN112631250B CN202011478933.3A CN202011478933A CN112631250B CN 112631250 B CN112631250 B CN 112631250B CN 202011478933 A CN202011478933 A CN 202011478933A CN 112631250 B CN112631250 B CN 112631250B
Authority
CN
China
Prior art keywords
data
fault
denoising
training
statistic
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202011478933.3A
Other languages
Chinese (zh)
Other versions
CN112631250A (en
Inventor
金佩薇
王浙超
曾九孙
姚燕
蔡晋辉
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China Jiliang University
Original Assignee
China Jiliang University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by China Jiliang University filed Critical China Jiliang University
Priority to CN202011478933.3A priority Critical patent/CN112631250B/en
Publication of CN112631250A publication Critical patent/CN112631250A/en
Application granted granted Critical
Publication of CN112631250B publication Critical patent/CN112631250B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B23/00—Testing or monitoring of control systems or parts thereof
    • G05B23/02—Electric testing or monitoring
    • G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
    • G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
    • G05B23/0275—Fault isolation and identification, e.g. classify fault; estimate cause or root of failure
    • G05B23/0281—Quantitative, e.g. mathematical distance; Clustering; Neural networks; Statistical analysis

Landscapes

  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Algebra (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Physics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Pure & Applied Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention discloses a fault isolation and identification method in a nonlinear process based on a denoising autoencoder. Acquiring training data and test data by using a sensor, and constructing statistics in a residual error space and a hidden feature space of a denoising self-encoder; calculating process monitoring statistics of the test data, and when the process monitoring statistics of the test data exceeds a process monitoring statistics control limit, generating a fault; a target function is established by introducing a regularization term, a fault amplitude is obtained based on the self-adaptive moment estimation solution, and the fault of each sensor is judged by utilizing the fault amplitude, so that the isolation and identification of the fault are realized. The invention can meet the speed and precision requirements of large-scale process fault diagnosis in the industrial process and provides reliable and effective technical support for industrial production process control.

Description

Fault isolation and identification method in nonlinear process based on denoising autoencoder
Technical Field
The invention belongs to a fault data processing method in the field of fault isolation and identification in an industrial nonlinear process, and particularly relates to a fault isolation and identification method in a nonlinear process based on a denoising self-encoder (DAE).
Background
The development of scientific technology makes modern industrial system structure more and more complex, and the safety and reliability of the system can be effectively ensured by a data-driven process monitoring method, wherein the research of multivariate statistical analysis methods such as Principal Component Analysis (PCA), kernel PCA, typical correlation analysis and the like is mainly used. However, although the conventional method can achieve fault detection, the development of industrial engineering increases the number of variables and samples, and the conventional method is limited by problems such as "tailing effect" in processing a nonlinear industrial process.
In recent years, deep learning methods, especially self-encoders (AEs), have been found to be effective in achieving nonlinear process monitoring. The self-encoder maps input nonlinear data to a low-dimensional space (encoding layer) through a multilayer neural network, then fits the mapped low-dimensional space data to approximate training data (decoding layer) by using a reconstruction objective function, and constructs monitoring statistics according to changes occurring in a hidden feature space and changes of a reconstruction error space. While AE enables industrial process monitoring, most approaches focus only on fault detection, thereby ignoring fault isolation and identification issues.
Disclosure of Invention
In order to solve the technical problems in the background art, the invention provides a fault isolation and identification method in a nonlinear process based on a Denoising Autoencoder (DAE), which introduces l into an objective function1And obtaining a sparse vector matrix by using the norm, and estimating a fault amplitude by using an ADAM algorithm to realize the separation of fault variables. The method can detect the fault occurrence time, isolate the fault and determine the fault variable through the nonlinear fault reconstruction technology, and has important significance for promoting the process industrial knowledge automation and the development of industrial big data technology.
The invention introduces l1Norm and other sparse characteristic constraints are used for obtaining a sparse contribution diagram, so that the problems of 'tailing effect' and the like in the traditional contribution diagram method are solved, main variables causing faults are identified, fault isolation is realized, and fault diagnosis in a large-scale process is facilitated.
The method of the invention can monitor the statistic H through the process constructed in the DAE residual error space and the hidden feature space2And R, fault diagnosis in the non-linear industrial process is realized, the reliability and safety of the non-industrial production process are realized, and the fault isolation and identification of the non-linear industrial process are improved. The method is less influenced by the tailing effect, and has important practical value for realizing fault isolation and identification in the nonlinear large-scale industrial production process.
The technical scheme adopted by the invention is as follows:
step 1, collecting real data in the operation process of industrial equipment under a normal working condition as training data through a distributed sensor, and collecting data in the operation process of the industrial equipment under a working condition to be tested as test data through the distributed sensor; obtaining damaged training data by adding noise to normal data or randomly discarding part of data in the normal data, and training a denoising autoencoder by using the damaged training data; then processing the test data by using the trained denoising autoencoder, wherein the denoising autoencoder outputs the statistic of the test data and the control limit thereof:
step 2, when the statistic of the test data output by the denoising autoencoder exceeds the control limit of the statistic of the training data, determining that a fault is monitored, wherein the test data is fault data, otherwise, determining that the fault is not monitored, and determining that the test data is normal data;
step 3, adding fault assignment into the fault data to obtain data after fault influence is eliminated, and eliminating the influence of the fault on each variable by introducing the fault assignment into the fault data to realize fault isolation so that the process variable with the fault in the test data reaches a normal state;
the variable refers to the detection value of the distributed sensor.
Step 4, taking the statistic of the test data and the training data as the process monitoring statistic, constructing an optimization objective function based on the process monitoring statistic, and adding l into the optimization objective function1The norm constraint is used for accurately positioning the main fault variable, so that the normal data are fitted with the training data after being reconstructed by the denoising autoencoder, and the fault data are separated from the training data after being reconstructed by the denoising autoencoder, thereby facilitating fault isolation;
and 5, solving the optimization objective function by adopting an adaptive matrix estimation (ADAM) algorithm, solving to obtain an optimal fault amplitude delta x, and realizing fault isolation and identification by utilizing the optimal fault amplitude delta x to finish the fault isolation and identification in the production process of the industrial equipment.
In the step 1, process monitoring is specifically performed according to a denoising autoencoder:
step 1.1: carrying out normalization processing on the damaged training data to obtain training data with zero mean and unit variance;
step 1.2: training a denoising self-encoder (DAE) by using the training data in the step 1.1:
the denoising self-encoder comprises an encoding process and a decoding process, training data are mapped to a hidden feature space through the encoding process of the denoising self-encoder, the denoising self-encoder decodes the mapping of the training data in the hidden feature space to obtain reconstructed data, then the reconstructed data and the training data are compared, and the denoising self-encoder is continuously optimized to enable the denoising self-encoder to learn the features of the training data;
step 1.3: normalizing the data sample in the test data to obtain the test data sample
Figure BDA0002836775190000021
Denoising autoencoder calculation test data sample obtained by training by using training data
Figure BDA0002836775190000022
H of (A) to (B)2Statistics and R statistics, where H2Statistical representation of test data samples
Figure BDA0002836775190000023
Mapping the change in the hidden feature space in the hidden layer of the denoising self-encoder through the denoising self-encoder, wherein the R statistic represents the change of a reconstruction error space caused by reconstructing variables and input variables by using the denoising self-encoder according to the hidden layer information;
H2the statistics are calculated as:
H2＝hTh
the R statistic was calculated as:
R＝eTe：
h represents low-dimensional data of a hidden space after dimension reduction of the denoising autoencoder, and e represents a difference value between reconstructed data of the denoising autoencoder and input data.
Detailed description of the invention2And the confidence limits of the R statistic are given by the kernel density estimate at the significance a. The denoise self-encoder outputs statistics of the test data and the reconstruction data together.
The data sample refers to a set of all variables collected at each moment in the test data.
In the step 3, each data sample of the fault data is input into the denoising autoencoder to judge whether a fault occurs, if the fault is monitored, a single data sample X of the fault data is obtained according to the following formula if the fault occursfDecomposing and subtracting the fault amplitude delta X to obtain data X after eliminating fault influence*Single data sample X of fault datafDecomposed into data X after eliminating fault influence*And fault amplitude Δ x:
X*＝Xf-Δx。
and the fault amplitude delta x is not zero, and an objective function is subsequently established to solve to obtain an estimated value of the fault amplitude, so that the isolation of the fault variable is realized.
In the step 4, an optimal problem based on process monitoring statistics is constructed, and process monitoring statistical data H is subjected to fault2And the influence of R when the statistic data of the test data exceeds H of the training data2And (3) establishing an optimized objective function of the following formula to estimate the delta x by the control limit of the statistic:
Figure BDA0002836775190000031
wherein the content of the first and second substances,
Figure BDA0002836775190000032
a recursion quantity representing a magnitude of the fault; λ represents an item weight; w represents the weight of the de-noised self-encoder; | Δ x | non-woven phosphor1Is the fault amplitude Deltax1Norm, σ, representing the sigmod activation function;
most process faults (especially early stages) usually affect only a few variables by introducing l1The norm obtains a sparse fault vector matrix delta x, which is beneficial to realizing the isolation and identification of faults;
when the statistic data of the test data exceeds the control limit of the R statistic of the training data, establishing an optimized objective function of the following formula to estimate the delta x:
Figure BDA0002836775190000033
wherein (X)f- Δ X)' is data X after eliminating the effect of a failure*T represents the matrix transpose, m represents the total number of data samples in the test data, | Δ x | survival2L being Δ x2And (4) norm.
The step 5 specifically comprises the following steps: when the fault amplitude delta x has elements which are not zero, the distributed sensors corresponding to the elements which are not zero in the fault amplitude delta x are in fault at the moment corresponding to the data samples, and therefore the fault conditions of all the distributed sensors are isolated and identified.
The ADAM enables the solution parameters to reduce the learning rate at the position with small gradient through first-order moment estimation and second-order moment estimation, and a descending path is searched more accurately, so that the oscillation at a saddle point is avoided, and the ADAM is enabled to be in a state1The sparse optimization problem has better performance.
In adaptive moment estimation (ADAM) algorithmic processing, α is the learning rate, which is typically set to 0.001, and ε is the accuracy tolerance, which is typically set to 10-8。
The invention trains the denoising self-encoder by adding noise to normal data or randomly discarding partial data to obtain a damaged data set, and establishes process monitoring statistics based on the change of a hidden feature space of the denoising self-encoder and the change of a reconstruction error space. The influence of the fault on each variable is eliminated by introducing the fault amplitude, and the fault process monitoring statistics is used for constructing an optimal problem for solving the fault amplitude. By introducing l into the optimization problem1The norm is used to obtain sparse solution, so as to solve the problems of 'tailing effect' and the like existing in the traditional contribution diagram method. And finally, solving the optimal problem by using ADAM, and obtaining a fault amplitude value which enables fault data to be closest to a normal value, so as to realize fault variable isolation and identification.
The invention introduces l into the traditional fault reconstruction objective function in the industrial process1And obtaining a sparse contribution diagram by the norm, isolating faults and determining fault variables by a nonlinear fault reconstruction technology, and improving the accuracy of fault positioning in the nonlinear industrial process.
Compared with the prior art, the invention has the following beneficial effects:
1. by introducing l in the reconstructed objective function1Obtaining a sparse vector matrix by using the norm so as to accurately position main variables causing faults, realize fault isolation and identification in a nonlinear process and realize innovation in the aspect of process industrial knowledge automation;
2. the method realizes fault isolation and fault variable identification through a nonlinear fault reconstruction technology, can detect the time of fault occurrence in a nonlinear industrial process, can judge the size and direction of the fault to realize fault isolation, and is favorable for engineering technicians to adjust and remedy the fault.
3. The method has the advantages that the influence of the trailing effect of the fault process variable on the normal process variable is greatly reduced by utilizing the sparse characteristic, the noise and interference resisting capability is good, and the method is very suitable for the nonlinear process.
4. The method fully utilizes process monitoring related knowledge and industrial production data, realizes more accurate fault isolation and identification of the nonlinear industrial process, and provides firm and reliable technical support for diagnosis of fault variables of the nonlinear industrial production process.
The invention can meet the speed and precision requirements of large-scale process fault diagnosis in the industrial process and provides reliable and effective technical support for industrial production process control.
Drawings
FIG. 1 is a schematic diagram of a blast furnace ironmaking process according to the present invention;
FIG. 2 is a DAE reconstruction result graph according to the present invention;
FIG. 3 is a DAE-based statistical monitoring result graph of the present invention;
fig. 4 is a diagram of the results of the DAE-based fault diagnosis of the present invention.
Detailed Description
The invention is further illustrated by the following figures and examples.
The embodiment and the implementation process of the complete method according to the invention are as follows:
taking the actual working process of a certain blast furnace ironmaking process in China as an example,
the actual working process of the coal mill in the coal-fired power plant is taken as an example, and the fault isolation and identification method of the process variable is described in detail based on the actual data recorded in the actual operation process.
As shown in fig. 1, in a blast furnace iron making process, iron-containing materials, fuels (pulverized coal and coke), and other materials are mixed in a certain ratio and then charged into the top of a blast furnace. Meanwhile, hot air heated by the hot blast stove is fed into an air inlet at the lower part of the blast furnace to be mixed with fuel, so that the combustion quality of the fuel is improved. At high temperature conditions in the furnace, the carbon reacts with the incoming air to form CO and H2Reducing gas (CO and H) at high temperature2) During the ascending process, the molten iron and the slag are generated after the reaction of heat transfer, reduction and the like with the raw materials. After the iron-making blast furnace breaks down, the staff need in time make adjustment and remedy to the trouble, not only need detect the point in time that the trouble takes place, still need keep apart the trouble, judge the size and the direction of trouble. Therefore, the method not only can detect the time of the fault occurrence, but also can isolate the fault and determine the fault variable through a nonlinear fault reconstruction technology.
TABLE 1
Figure BDA0002836775190000051
The blast furnace ironmaking process involves 8 process variables (denoted as mu) in total1To mu8) The specific labels and variable names are shown in table 1. The data set consists of 2500 sample points, wherein the last 410 sample points are blast furnace suspension fault sample data, the data acquisition time interval is 20 minutes, the first 1500 sample points of the data set are used as a training set, and the last 1000 sample points are used as a test set. The failure causes the blast quantity mu1With the pressure in the furnace mu3Seriously decreases, resulting in great changes in the gas component content and coal injection amount of the top-blown gas
In order to apply the fault isolation and identification method based on the denoising autoencoder to monitor, separate and identify the faults in the blast furnace ironmaking industrial process, the following steps are formulated:
step 1, acquiring real data in the operation process of industrial equipment under normal working conditions as training data through a distributed sensor, adding noise into the normal data or randomly discarding part of data in the normal data to obtain damaged training data, and training a denoising autoencoder by using the damaged training data; acquiring data in the operation process of industrial equipment under a working condition to be tested as test data through a distributed sensor; when the process monitoring statistic of the test data monitored by the denoising autoencoder exceeds the process monitoring statistic control limit of the training data, determining that a fault is monitored, otherwise, determining that the fault is not monitored;
in order to detect the failure, firstly, noise is added to 1500 normal data or part of the normal data is randomly discarded to obtain damaged data, and then the self-encoder is trained by using the obtained damaged data set. A fault detection confidence level is set to 99%.
And 2, if the fault is monitored, the influence of the fault on each variable is eliminated by introducing fault assignment into fault data, so that fault isolation is realized, and the process variable with the fault in the test data reaches a normal state. The results of 2500 data samples reconstructed by the DAE are shown in fig. 2, and the implementation represents the training data, and the dotted line represents the reconstructed data obtained by the DAE. Normal data is fit to training data after DAE, while failure data is reconstructed and separated from training data.
Step 3, constructing an optimal problem based on process monitoring statistics, and introducing l into an optimal objective function1The norm constraint accurately positions main fault variables, normal data are fitted with training data after being reconstructed by the denoising autoencoder, and fault data are separated from the training data after being reconstructed by the denoising autoencoder, so that fault isolation is facilitated. DAE implements separation of normal data from fault data, and in order to determine when a fault occurs, H is used2And R statistics, and carrying out process monitoring on the statistical information, as shown in figure 3, wherein the condition of violating the confidence limit appears from the 2090 point, and the condition is consistent with the condition of actually generating the suspension fault of the blast furnace.
And 4, solving the optimization objective function mentioned in the step 3 by adopting an adaptive matrix estimation (ADAM) algorithm, taking the obtained optimal estimation vector as an optimal fault amplitude, and realizing fault isolation and identification by utilizing the optimal fault amplitude to finish the fault isolation and identification in the production process of the industrial equipment.
After the trouble takes place, in order to make adjustment and remedy to the trouble in time for the staff, not only need detect the time point that the trouble takes place, still need carry out fault isolation, judge the size and the direction of trouble. After suspension failure [2140, 2340]The results of the fault diagnosis in the sample point interval are shown in fig. 4. Blast amount mu caused by failure1With the pressure in the furnace mu3A severe drop, after which the fuel concentration in the hot blast is increased due to the drop in the blast volume, resulting in the sensor μ detecting the coal injection amount4The numerical value is increased, the pressure in the furnace is reduced to suddenly reduce the reaction of CO and the iron-containing raw materials, and CO in the top-blown gas is caused2Sudden decrease in concentration of CO and H2The content increases.
In summary, the fault isolation and identification method in the nonlinear process based on the Denoising Autoencoder (DAE) can complete fault monitoring of the nonlinear industrial process, realize isolation and identification of the process variable of the fault, and effectively improve the sensitivity of fault monitoring and the accuracy of fault information positioning.

Claims (3)

1. A fault isolation and identification method based on a denoising autoencoder is characterized by comprising the following steps: the method comprises the following steps:
step 1, collecting real data in the operation process of industrial equipment under a normal working condition as training data through a distributed sensor, and collecting data in the operation process of the industrial equipment under a working condition to be tested as test data through the distributed sensor; obtaining damaged training data by adding noise to normal data or randomly discarding part of data in the normal data, and training a denoising autoencoder by using the damaged training data; then processing the test data by using the trained denoising autoencoder, wherein the denoising autoencoder outputs the statistic of the test data and the control limit thereof:
step 2, when the statistic of the test data output by the denoising autoencoder exceeds the control limit of the statistic of the training data, determining that a fault is monitored, wherein the test data is fault data, otherwise, determining that the fault is not monitored, and determining that the test data is normal data;
step 3, adding fault assignment into the fault data to obtain data after fault influence is eliminated, and eliminating the influence of the fault on each variable by introducing the fault assignment into the fault data to realize fault isolation so that the process variable with the fault in the test data reaches a normal state;
step 4, taking the statistic of the test data and the training data as process monitoring statistic, constructing an optimization objective function based on the process monitoring statistic, and fitting the normal data after the normal data is reconstructed by the denoising self-encoder with the training data;
in step 4, when the statistic data of the test data exceeds the H of the training data2And (3) establishing an optimized objective function of the following formula to estimate the delta x by the control limit of the statistic quantity:
Figure FDA0003629034460000012
wherein the content of the first and second substances,
Figure FDA0003629034460000013
a recursion quantity representing a magnitude of the fault; λ represents an item weight; w represents the weight of a denoising autocoder; | Δ x | non-woven phosphor1Is the fault amplitude Deltax1Norm,. σ denotes the sigmod activation function;
when the statistic data of the test data exceeds the control limit of the R statistic of the training data, establishing an optimized objective function of the following formula to estimate the delta x:
Figure FDA0003629034460000011
wherein (X)f- Δ X)' is data X after eliminating the effect of a failure*T denotes a matrix transpose, m denotes data in the test dataTotal number of samples, | Δ x | | non-woven phosphor2L being Δ x2Norm, XfA single data sample that is fault data;
step 5, solving the optimization objective function, solving to obtain an optimal fault amplitude delta x, realizing fault isolation and identification by using the optimal fault amplitude delta x, and completing fault isolation and identification in the production process of the industrial equipment;
in the step 3, if the data is fault data, a single data sample X of the fault data is obtained according to the following formulafDecomposing and subtracting the fault amplitude delta X to obtain data X after eliminating fault influence*：
X*＝Xf-Δx。
2. The method for fault isolation and identification based on the denoising autoencoder as claimed in claim 1, wherein: in the step 1, process monitoring is specifically performed according to a denoising autoencoder:
step 1.1: carrying out normalization processing on the damaged training data to obtain training data with zero mean and unit variance;
step 1.2: training a denoising self-encoder (DAE) by using the training data in the step 1.1:
the denoising self-encoder comprises an encoding process and a decoding process, the training data are mapped to the hidden characteristic space through the encoding process of the denoising self-encoder, the denoising self-encoder decodes the mapping of the training data in the hidden characteristic space to obtain reconstructed data, then the reconstructed data and the training data are compared, the denoising self-encoder is continuously optimized, and the denoising self-encoder learns the characteristics of the training data;
step 1.3: normalizing the data sample in the test data to obtain the test data sample
Figure FDA0003629034460000021
Denoising autoencoder calculation test data sample obtained by training by using training data
Figure FDA0003629034460000022
H of (A) to (B)2Statistics and R statistics;
H2the statistics are calculated as:
H2＝hTh
the R statistic is calculated as:
R＝eTe：
h represents low-dimensional data of a hidden space after dimension reduction of the denoising autoencoder, and e represents a difference value between reconstructed data of the denoising autoencoder and input data.
3. The method for fault isolation and identification based on the denoising autoencoder as claimed in claim 1, wherein: the step 5 specifically comprises the following steps: the obtained fault amplitude delta x is an n-dimensional vector, n is the total number of distributed sensors which acquire data at each moment, namely the number of data values in a data sample, and the distributed sensors corresponding to the non-zero elements in the fault amplitude delta x have faults at the moment corresponding to the data sample, so that the fault conditions of all the distributed sensors are isolated and identified.
CN202011478933.3A 2020-12-15 2020-12-15 Fault isolation and identification method in nonlinear process based on denoising autoencoder Active CN112631250B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011478933.3A CN112631250B (en) 2020-12-15 2020-12-15 Fault isolation and identification method in nonlinear process based on denoising autoencoder

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011478933.3A CN112631250B (en) 2020-12-15 2020-12-15 Fault isolation and identification method in nonlinear process based on denoising autoencoder

Publications (2)

Publication Number Publication Date
CN112631250A CN112631250A (en) 2021-04-09
CN112631250B true CN112631250B (en) 2022-06-24

Family

ID=75313132

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011478933.3A Active CN112631250B (en) 2020-12-15 2020-12-15 Fault isolation and identification method in nonlinear process based on denoising autoencoder

Country Status (1)

Country Link
CN (1) CN112631250B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113468751B (en) * 2021-07-05 2022-12-27 河南中烟工业有限责任公司 Recursion Lasso-based flowmeter anomaly online monitoring method and system and storage medium

Family Cites Families (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1899890B1 (en) * 2005-05-02 2012-09-05 Ultraspect Ltd. System for image reconstruction
CN105548832B (en) * 2015-12-10 2018-09-21 国网四川省电力公司电力科学研究院 High voltage power cable fault recognition method
US10216591B1 (en) * 2016-06-30 2019-02-26 EMC IP Holding Company LLC Method and apparatus of a profiling algorithm to quickly detect faulty disks/HBA to avoid application disruptions and higher latencies
US11205103B2 (en) * 2016-12-09 2021-12-21 The Research Foundation for the State University Semisupervised autoencoder for sentiment analysis
CN108594790B (en) * 2018-04-11 2019-12-10 浙江大学 Fault detection and separation method based on structured sparse principal component analysis
CN108871762A (en) * 2018-06-29 2018-11-23 广东工业大学 A kind of gearbox of wind turbine method for diagnosing faults
CN109242097B (en) * 2018-07-12 2020-09-25 上海交通大学 Visual representation learning system and method for unsupervised learning
CN109408552B (en) * 2018-08-08 2022-01-04 南京航空航天大学 Fault monitoring and identifying method for civil aircraft system based on LSTM-AE deep learning framework
CN109186973B (en) * 2018-08-20 2020-03-31 东南大学 Mechanical fault diagnosis method of unsupervised deep learning network
CN109870628B (en) * 2018-08-31 2020-12-04 国网江苏省电力有限公司苏州供电分公司 Fault line identification method for multi-terminal flexible direct current transmission system
CN110286279B (en) * 2019-06-05 2021-03-16 武汉大学 Power electronic circuit fault diagnosis method based on extreme tree and stack type sparse self-coding algorithm
CN110147648B (en) * 2019-06-20 2022-06-17 浙江大学 Automobile sensor fault detection method based on independent component analysis and sparse denoising self-encoder
CN110705722A (en) * 2019-09-12 2020-01-17 华中科技大学 Diagnostic model for industrial equipment fault diagnosis and construction method and application thereof
CN111310830B (en) * 2020-02-17 2023-10-10 湖北工业大学 Blocking fault diagnosis system and method for combine harvester
CN112051839A (en) * 2020-09-17 2020-12-08 中国计量大学 Process monitoring and fault diagnosis method based on tree structure sparsity

Also Published As

Publication number Publication date
CN112631250A (en) 2021-04-09

Similar Documents

Publication Publication Date Title
CN112257530B (en) Rolling bearing fault diagnosis method based on blind signal separation and support vector machine
CN104793606B (en) Industrial method for diagnosing faults based on improved KPCA and HMM
CN108549908B (en) Chemical process fault detection method based on multi-sampling probability kernel principal component model
CN112631250B (en) Fault isolation and identification method in nonlinear process based on denoising autoencoder
CN111752147B (en) Multi-working-condition process monitoring method with continuous learning capability and improved PCA (principal component analysis)
CN109270907A (en) A kind of process monitoring and method for diagnosing faults based on the decomposition of stratified probability density
CN111367959A (en) Zero-time-lag nonlinear expansion Granger causal analysis method
CN111076934A (en) Method for diagnosing potential fault of bearing based on S transformation
CN112395684A (en) Intelligent fault diagnosis method for high-speed train running part system
CN113703422B (en) Gas turbine pneumatic actuator fault diagnosis method based on feature analysis processing
CN104499001A (en) Aluminum electrolysis cell condition diagnosing method based on sub-feature space optimization relative matrix
CN111188761A (en) Monitoring method for pump equipment based on Fourier-CVA model
CN108709426B (en) Sintering machine air leakage fault online diagnosis method based on frequency spectrum characteristic bilateral detection method
CN114200914A (en) MW-OCCA-based quality-related early fault detection method
CN114707424B (en) Chemical process soft measurement method based on quality-related slow characteristic analysis algorithm
CN116339275A (en) Multi-scale process fault detection method based on full-structure dynamic autoregressive hidden variable model
CN111931574B (en) Robust fault diagnosis method for pneumatic regulating valve
CN115167364A (en) Early fault detection method based on probability transformation and statistical characteristic analysis
CN113111752B (en) Rolling mill fault diagnosis method for sample imbalance enhanced extended depth confidence network
CN112784744B (en) Mechanical component vibration signal preprocessing method with missing value
CN114581699A (en) Transformer state evaluation method based on deep learning model in consideration of multi-source information
CN112051839A (en) Process monitoring and fault diagnosis method based on tree structure sparsity
CN108508866B (en) Million kilowatt unit fault variable identification method based on sparse relative discriminant analysis
CN113325811B (en) Online industrial process anomaly detection method based on memory and forgetting strategy
CN114185321B (en) Electric actuator fault diagnosis method for improving multi-classification twin support vector machine

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant